Sync Confluence open model registry

README.md

@@ -0,0 +1,37 @@
+---
+license: mit
+tags:
+- models
+- retrieval
+- reranking
+- llm
+- healthcare
+- open-science
+---
+
+# Confluence Open Model Registry
+
+This repository does not ship model weights. It is a transparent registry of external open models that are candidates for Confluence research workflows.
+
+The current focus is narrow on purpose:
+
+- retrieval
+- reranking
+- structured research assistance
+
+These models are not designated here for autonomous clinical use.
+
+## Workflow Roles
+
+- embeddings for dataset, paper, and protocol retrieval
+- reranking for higher-precision search
+- instruction-tuned reasoning models for non-clinical extraction and summarization
+
+## Selection Rule
+
+Models enter the Confluence workflow only after benchmark comparison against explicit tasks and governance constraints.
+
+## Files
+
+- `registry.json`: candidate model list
+- `benchmark_policy.md`: how Confluence should evaluate and select open models

benchmark_policy.md

@@ -0,0 +1,27 @@
+# Benchmark Policy
+
+Confluence should benchmark open models before they are integrated into production research workflows.
+
+## Required Evaluation Axes
+
+- retrieval relevance
+- reranking accuracy
+- extraction precision
+- hallucination rate
+- citation faithfulness
+- latency
+- memory footprint
+- governance fit
+
+## Safety Rule
+
+No model in this registry is a treatment authority. Models are support tools for evidence retrieval, organization, and non-clinical reasoning.
+
+## Promotion Rule
+
+A candidate may move from `benchmark` to `approved` only if it:
+
+- improves a defined workflow metric
+- stays within acceptable cost and latency bounds
+- does not degrade citation faithfulness
+- does not encourage unsupported clinical claims

registry.json

@@ -0,0 +1,63 @@
+{
+  "version": "0.1.0",
+  "roles": [
+    {
+      "role": "embedding",
+      "recommended_for": [
+        "paper retrieval",
+        "dataset retrieval",
+        "protocol retrieval"
+      ],
+      "candidates": [
+        {
+          "repo_id": "BAAI/bge-m3",
+          "status": "benchmark",
+          "notes": "Multilingual retrieval candidate with long-context support."
+        }
+      ]
+    },
+    {
+      "role": "reranker",
+      "recommended_for": [
+        "top-k reranking",
+        "evidence ordering"
+      ],
+      "candidates": [
+        {
+          "repo_id": "BAAI/bge-reranker-v2-m3",
+          "status": "benchmark",
+          "notes": "Multilingual reranker candidate for biomedical search refinement."
+        }
+      ]
+    },
+    {
+      "role": "instruct_small",
+      "recommended_for": [
+        "structured extraction",
+        "research summarization",
+        "non-clinical reasoning"
+      ],
+      "candidates": [
+        {
+          "repo_id": "google/gemma-3-4b-it",
+          "status": "benchmark",
+          "notes": "Smaller instruction-tuned candidate for cost-sensitive workflows."
+        }
+      ]
+    },
+    {
+      "role": "instruct_medium",
+      "recommended_for": [
+        "higher-complexity synthesis",
+        "longer-context research assistance"
+      ],
+      "candidates": [
+        {
+          "repo_id": "google/gemma-3-12b-it",
+          "status": "benchmark",
+          "notes": "Medium-size reasoning candidate; benchmark before adoption."
+        }
+      ]
+    }
+  ]
+}